Privacy leakage detection method based on parallel deep neural network

Abstract

The embodiment of the present application provides a privacy leakage detection method based on a parallel deep neural network. In the embodiment of the present application, sensitive privacy features are first extracted from each privacy sample data of the obtained privacy data set to obtain a sensitive privacy feature vector corresponding to the privacy sample data; then, the N CPU cores on the device train the corresponding DNN model in parallel based on the sensitive privacy feature vector corresponding to each privacy sample data, so that when any detection device needs to detect any target network flow, the trained DNN model is used to detect whether the target network flow has privacy leakage, thereby realizing the detection of privacy leakage of IoT devices and ensuring the information security of IoT devices. In this embodiment, multiple DNN models are trained in parallel by N CPU cores, so that the model parameters between the DNN models will not affect each other, avoiding coupling between DNN models, and improving the performance of each DNN model.

Description

Technical Field

[0001] The present application relates to the field of information security technology, and in particular to a privacy leakage detection method based on a parallel deep neural network. Background Art

[0002] With the rapid development of the Internet of Things, the number of IoT devices (such as smart homes, smart watches, etc.) is also growing. Attackers (such as hackers) have begun to pay attention to and use these IoT devices to launch attacks to obtain sensitive information or private information from IoT devices, which will cause the leakage of sensitive information or private information and affect the information security of IoT devices. Therefore, how to detect whether there is privacy leakage in IoT devices is a technical problem that needs to be solved urgently. Summary of the invention

[0003] In view of this, the present application provides a privacy leakage detection method based on a parallel deep neural network to realize the detection of privacy leakage of IoT devices and ensure the information security of IoT devices.

[0004] The embodiment of the present application provides a privacy leakage detection method based on a parallel deep neural network, which is applied to an electronic device. The method includes:

[0005] Obtain a privacy data set; the privacy data set includes privacy sample data corresponding to multiple network flows respectively; any network flow is initiated when a business function with privacy data is triggered;

[0006] Extracting at least one sensitive privacy feature under a sensitive information category from each privacy sample data of the privacy data set to obtain a sensitive privacy feature vector corresponding to the privacy sample data;

[0007] Based on the sensitive privacy feature vector corresponding to each privacy sample data, control N central processing unit (CPU) cores on the device to train the corresponding deep neural network (DNN) model in parallel; in the DNN model corresponding to any CPU core, the number of nodes on the input layer corresponds to the number of eigenvalues ​​in the sensitive privacy feature vector, and the number of nodes on the output layer corresponds to the number of sensitive information categories; different CPU cores use different optimizers for training the DNN model; N is greater than 1;

[0008] Among them, the DNN models corresponding to the N CPU cores are used. When any detection device needs to detect any target network flow, the detection device selects the DNN models that match the M CPU cores respectively from the DNN models corresponding to the N CPU cores based on the M idle CPU cores in the detection device, and the M CPU cores use the matching DNN models to perform IoT privacy leakage detection on the target network flow, and determine the target prediction result based on the M detection results; the target prediction result indicates whether there is privacy leakage in the target network flow, M is greater than 1 and less than or equal to N; when M is less than N, the DNN model matched by any CPU core is trained by a CPU core with the same architecture as the CPU core and has the highest model accuracy among all the remaining DNN models.

[0009] The embodiment of the present application also provides a privacy leakage detection device based on a parallel deep neural network, which is applied to an electronic device, and includes:

[0010] The acquisition module is used to obtain a private data set; the private data set includes private sample data corresponding to multiple network flows respectively; any network flow is initiated when a business function with private data is triggered;

[0011] An extraction module is used to extract at least one sensitive privacy feature under a sensitive information category from each privacy sample data in the privacy data set to obtain a sensitive privacy feature vector corresponding to the privacy sample data;

[0012] A training module, used to control N CPU cores on the device to train the corresponding DNN model in parallel based on the sensitive privacy feature vector corresponding to each privacy sample data; in the DNN model corresponding to any CPU core, the number of nodes on the input layer corresponds to the number of eigenvalues ​​in the sensitive privacy feature vector, and the number of nodes on the output layer corresponds to the number of sensitive information categories; different CPU cores use different optimizers for training the DNN model; N is greater than 1;

[0013] Among them, the DNN models corresponding to the N CPU cores are used. When any detection device needs to detect any target network flow, the detection device selects the DNN models that match the M CPU cores respectively from the DNN models corresponding to the N CPU cores based on the M idle CPU cores in the detection device, and the M CPU cores use the matching DNN models to perform IoT privacy leakage detection on the target network flow, and determine the target prediction result based on the M detection results; the target prediction result indicates whether there is privacy leakage in the target network flow, M is greater than 1 and less than or equal to N; when M is less than N, the DNN model matched by any CPU core is trained by a CPU core with the same architecture as the CPU core and has the highest model accuracy among all the remaining DNN models.

[0014] The present application also provides an electronic device, the electronic device comprising:

[0015] Processor; and

[0016] A computer-readable storage medium stores computer program instructions, which, when executed by a processor, cause the processor to execute the steps of the above method.

[0017] An embodiment of the present application further provides a computer-readable storage medium, on which computer program instructions are stored. When the computer program instructions are executed by a processor, the processor executes the steps in the above method.

[0018] It can be seen from the above technical scheme that in the embodiment of the present application, sensitive privacy features are first extracted from each privacy sample data of the obtained privacy data set to obtain a sensitive privacy feature vector corresponding to the privacy sample data; then, the N CPU cores on the device are used to train the corresponding DNN model in parallel based on the sensitive privacy feature vectors corresponding to each privacy sample data, so that when any detection device needs to detect any target network flow, the trained DNN model can be used to detect whether there is privacy leakage in the target network flow, thereby realizing the detection of privacy leakage of IoT devices and ensuring the information security of IoT devices.

[0019] Furthermore, in the embodiment of the present application, N CPU cores are used to train multiple DNN models in parallel, and each DNN model shares the same training data, as well as the same input layer parameters and output layer parameters, that is, the number of nodes configured on the input layer and the output layer. This not only prevents the model parameters of the DNN models from affecting each other and avoids coupling between DNN models, but also eliminates the need for complex integrated learning of the DNN models, thereby reducing the computational complexity of model training and shortening the model training time. It can also significantly reduce the number of parameters required for model training, thereby reducing the memory usage of the parameters required for model training. Based on this, different CPU cores in this embodiment use different optimizers for training DNN models, which can reduce the impact of the DNN model on the selection of a single optimizer and reduce the dependence between the DNN models, thereby effectively improving the performance of each DNN model, and thereby improving the accuracy of privacy leakage detection in the Internet of Things.

[0020] Furthermore, in an embodiment of the present application, all idle CPU cores in the detection device are utilized to perform privacy leakage detection on the target network flow based on a DNN model selected from the trained DNN models and matching each idle CPU core, so as to comprehensively determine whether there is a privacy leakage detection of the target network flow based on the detection results corresponding to each idle CPU core. This can not only make full use of the CPU core resources of the detection device to avoid resource waste, but also further improve the accuracy of privacy leakage detection of the network flow. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] The accompanying drawings, which are incorporated in the specification and constitute a part of this application, illustrate embodiments consistent with the application and, together with the description, serve to explain the principles of the application.

[0022] Figure 1 A flow chart of the method provided in an embodiment of the present application.

[0023] Figure 2 Another method flow chart provided for an embodiment of the present application.

[0024] Figure 3 Another method flow chart provided for an embodiment of the present application.

[0025] Figure 4 A schematic diagram of the device structure provided in an embodiment of the present application.

[0026] Figure 5 A schematic diagram of the structure of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0027] In order to enable those skilled in the art to better understand the technical solutions provided by the embodiments of the present application and to make the above-mentioned purposes, features and advantages of the embodiments of the present application more obvious and understandable, the technical solutions in the embodiments of the present application are further described in detail below in conjunction with the accompanying drawings.

[0028] See also Figure 1 , Figure 1 The method flow chart provided in the embodiment of the present application. The method is applied to an electronic device. As an embodiment, the electronic device may be, for example, a training server or an Internet of Things device, etc., which is not specifically limited in this embodiment. Figure 1 As shown, the process may include the following steps:

[0029] Step 101, obtaining a privacy data set; the privacy data set includes privacy sample data corresponding to a plurality of network flows.

[0030] In this embodiment, as an embodiment, the above-mentioned private data set has been pre-stored locally in the electronic device. Based on this, the step of obtaining the private data set can be implemented as follows: obtaining the private data set from the local electronic device. As another embodiment, the above-mentioned private data set is stored in other devices other than the electronic device. Based on this, the step of obtaining the private data set can be implemented as follows: obtaining the private data set from other devices storing the private data set.

[0031] In this embodiment, the above-mentioned privacy data set includes privacy sample data corresponding to multiple network flows respectively; wherein any network flow is initiated when a business function with privacy data is triggered. For example, as an embodiment, the network flow initiated when the business function is triggered can be obtained by actively triggering the business function containing privacy data on the IoT device. It should be noted that in this embodiment, there is no specific limitation on the number of IoT devices used to obtain network flows, for example, it can be one or more; wherein, if the electronic device is an IoT device, then the electronic device can also be used as an IoT device for obtaining network flows.

[0032] In this embodiment, the number of network flows used to constitute a privacy data set can be determined based on actual application requirements. Based on this, after obtaining multiple network flows whose number meets the actual application requirements, a corresponding privacy data set is constructed based on all the obtained network flows, and the privacy data set is stored in a designated storage medium for backup.

[0033] In this embodiment, the network flow refers to a network data flow based on an Internet of Things communication protocol (such as HTTP protocol, HTTPS protocol, etc.). This embodiment does not specifically limit the data type of the network flow. For example, as an embodiment, the data type of the network flow may include text type and video type, etc.

[0034] The above-mentioned business functions refer to the business service functions provided by the IoT devices. This embodiment does not specifically limit the business functions containing privacy data. For example, as an embodiment, the business functions containing privacy data may include but are not limited to user login business functions, location sharing business functions, payment transaction business functions, and video stream extraction business functions. Correspondingly, the data content of the network flow initiated when different business functions are triggered is different; for example, the data content of the network flow initiated when the user login business function is triggered may include user information and login key information, etc.; the data content of the network flow initiated when the location sharing business function is triggered may include device information and location information, etc.; this embodiment does not specifically limit this.

[0035] Step 102: extract at least one sensitive privacy feature under a sensitive information category from each privacy sample data in the privacy data set to obtain a sensitive privacy feature vector corresponding to the privacy sample data.

[0036] In this embodiment, multiple sensitive information categories can be pre-set based on actual application requirements, such as identity identification category, device identification category, data transmission information category, and key category, etc. Among them, sensitive information belonging to the identity identification category may include identity card information, facial recognition information, and fingerprint recognition information, etc.; sensitive information belonging to the device identification category may include device name, device IP address, and port number used by the device, etc.; sensitive information belonging to the data transmission information category may include data transmission protocol type, data transmission amount, and duration of data transmission, etc.; sensitive information belonging to the key category may include login key information, etc.; this embodiment does not specifically limit it.

[0037] In this embodiment, any network flow includes sensitive information under at least one sensitive information category; based on this, for each private sample data in the private data set, at least one sensitive privacy feature under the sensitive information category can be extracted from the private sample data to obtain a sensitive privacy feature vector corresponding to the private sample data for subsequent model training. As for how to extract at least one sensitive privacy feature under the sensitive information category from the private sample data to obtain a sensitive privacy feature vector corresponding to the private sample data, an example will be given below and will not be repeated here.

[0038] Step 103: Based on the sensitive privacy feature vectors corresponding to each privacy sample data, control the N CPU cores on the device to train the corresponding DNN models in parallel; wherein, the DNN models corresponding to the N CPU cores are used for, when any detection device needs to detect any target network flow, the detection device selects the DNN models that match the M CPU cores respectively from the DNN models corresponding to the N CPU cores based on the M idle CPU cores in the detection device, and the M CPU cores respectively use the matching DNN models to perform IoT privacy leakage detection on the target network flow, and determine the target prediction result based on the M detection results; the target prediction result indicates whether there is privacy leakage in the target network flow, and M is greater than 1 and less than or equal to N.

[0039] In this embodiment, this device, i.e., the above-mentioned electronic device, includes N CPU cores, where N is greater than 1. Based on this, after obtaining the sensitive privacy feature vector corresponding to each privacy sample data in the privacy data set, the sensitive privacy feature vector corresponding to each privacy sample data can be used as training data for model training, and the N CPU cores on this device are controlled to train the corresponding DNN model in parallel based on the training data. As for how to train the DNN model based on the training data, this embodiment does not make specific restrictions, and the model training method can be determined based on actual needs to train the DNN model based on the model training method.

[0040] In this embodiment, the DNN model corresponding to any CPU core includes at least an input layer, a hidden layer, and an output layer. Among them, in the DNN model corresponding to any CPU core, the number of nodes on the input layer corresponds to the number of eigenvalues ​​in the sensitive privacy feature vector; here, the node on the input layer refers to the input neuron on the input layer, which is used to receive the input data of the model. Moreover, in this embodiment, the number of eigenvalues ​​in the sensitive privacy feature vector corresponding to each privacy sample data is the same; based on this, as an embodiment, the number of nodes on the input layer corresponds to the number of eigenvalues ​​in the sensitive privacy feature vector, for example, it means that the number of nodes on the input layer is the same as the number of eigenvalues ​​in the sensitive privacy feature vector, and for example, it means that the number of nodes on the input layer is greater than the number of eigenvalues ​​in the sensitive privacy feature vector, and this embodiment does not specifically limit this.

[0041] In the DNN model corresponding to any CPU core, the number of nodes on the output layer corresponds to the number of sensitive information categories; here, the nodes on the output layer refer to the output neurons on the output layer, which are used to output the output results of the model; the number of sensitive information categories refers to the number of pre-set sensitive information categories described in the above step 102. As an embodiment, the number of nodes on the output layer corresponds to the number of sensitive information categories, for example, the number of nodes on the output layer is the same as the number of sensitive information categories, or the number of nodes on the output layer is greater than the number of sensitive information categories, which is not specifically limited in this embodiment.

[0042] In this embodiment, in order to reduce the impact of the model on the selection of a single optimizer and improve the robustness and generalization ability of the model, different CPU cores use different optimizers for training the DNN model. Here, the optimizer refers to a module used to adjust model parameters to optimize the model; for example, the optimizer used in this embodiment may include but is not limited to Adam, Nadam, Adabelief, and Adamax optimizers.

[0043] In this embodiment, after the N CPU cores on the device are controlled to train the corresponding DNN model in parallel, when any detection device needs to detect any target network flow, the DNN model corresponding to the trained N CPU cores can be used to detect whether there is privacy leakage in the target network flow. Here, the detection device can be the same device as the device or a different device from the device, which is not specifically limited in this embodiment.

[0044] Specifically, this embodiment first obtains M idle CPU cores in the detection device, where M is greater than 1 and less than or equal to N; then, the detection device selects DNN models that match the M CPU cores respectively from the DNN models corresponding to the N CPU cores based on the M CPU cores; finally, the M CPU cores use the matching DNN models to perform IoT privacy leakage detection on the above-mentioned target network flow to obtain M detection results, so as to determine the target prediction result based on the M detection results, wherein the target prediction result indicates whether there is privacy leakage in the target network flow, thereby realizing privacy leakage detection of the target network flow.

[0045] In this embodiment, as an embodiment, the above-mentioned obtaining of the M idle CPU cores in the detection device can be specifically implemented as follows: checking whether there are idle CPU cores on the detection device through a resource monitoring module configured on the detection device for monitoring the usage of the CPU cores, and if so, obtaining the M idle CPU cores in the detection device.

[0046] In this embodiment, at least two of the N CPU cores have different architectures. Here, the architecture of the CPU core may refer to the instruction set architecture of the CPU core, which is not specifically limited in this embodiment; the instruction set architecture may include, for example, x86 and Advanced RISC Machine (ARM).

[0047] Based on the above description, in order to better utilize the performance of the DNN model trained by N CPU cores and further improve the accuracy of privacy leakage detection, in actual application, this embodiment can try to utilize DNN models with high model accuracy, and try to use CPU cores with the same architecture as the CPU core that trained the DNN model to use the DNN model for privacy leakage detection. Based on this, when M is equal to N, the DNN model matched by any CPU core is as far as possible a DNN model trained by a CPU core with the same architecture as the CPU core; and when M is less than N, the DNN model matched by any CPU core is trained by a CPU core with the same architecture as the CPU core and has the highest model accuracy among all the remaining DNN models.

[0048] As for how to select the DNN models that match the M CPU cores from the DNN models corresponding to the N CPU cores, and how to determine the target prediction results based on the M detection results, the following will give examples and will not be elaborated here.

[0049] So far, completed Figure 1 The process shown.

[0050] pass Figure 1 It can be seen from the shown process that in the embodiment of the present application, sensitive privacy features are first extracted from each privacy sample data of the obtained privacy data set to obtain a sensitive privacy feature vector corresponding to the privacy sample data; then, the N CPU cores on the device train the corresponding DNN model in parallel based on the sensitive privacy feature vectors corresponding to each privacy sample data, so that when any detection device needs to detect any target network flow, the trained DNN model can be used to detect whether there is privacy leakage in the target network flow, thereby realizing the detection of privacy leakage of IoT devices and ensuring the information security of IoT devices.

[0051] Furthermore, in the embodiment of the present application, N CPU cores are used to train multiple DNN models in parallel, and each DNN model shares the same training data, as well as the same input layer parameters and output layer parameters, that is, the number of nodes configured on the input layer and the output layer. This not only prevents the model parameters of the DNN models from affecting each other and avoids coupling between DNN models, but also eliminates the need for complex integrated learning of the DNN models, thereby reducing the computational complexity of model training and shortening the model training time. It can also significantly reduce the number of parameters required for model training, thereby reducing the memory usage of the parameters required for model training. Based on this, different CPU cores in this embodiment use different optimizers for training DNN models, which can reduce the impact of the DNN model on the selection of a single optimizer and reduce the dependence between the DNN models, thereby effectively improving the performance of each DNN model, and thereby improving the accuracy of privacy leakage detection in the Internet of Things.

[0052] Furthermore, in an embodiment of the present application, all idle CPU cores in the detection device are utilized to perform privacy leakage detection on the target network flow based on a DNN model selected from the trained DNN models and matching each idle CPU core, so as to comprehensively determine whether there is a privacy leakage detection of the target network flow based on the detection results corresponding to each idle CPU core. This can not only make full use of the CPU core resources of the detection device to avoid resource waste, but also further improve the accuracy of privacy leakage detection of the network flow.

[0053] The following describes how to extract at least one sensitive privacy feature under the sensitive information category from each privacy sample data in the privacy data set in the above step 102 to obtain a sensitive privacy feature vector corresponding to the privacy sample data:

[0054] As an embodiment, the above method extracts at least one sensitive privacy feature under the sensitive information category from each privacy sample data in the privacy data set to obtain the sensitive privacy feature vector corresponding to the privacy sample data. There are many ways to implement it in specific implementation; for example, see Figure 2 As shown, the specific implementation may include the following steps:

[0055] Step 201: for each privacy sample data in the privacy data set, extract at least one sensitive privacy feature under the sensitive information category from the privacy sample data, and convert the non-numerical data in each extracted sensitive privacy feature into numerical data to obtain each sensitive privacy feature value corresponding to the privacy sample data.

[0056] In this embodiment, as an embodiment, this step 201 can be implemented specifically as follows: first, feature extraction is performed on the privacy sample data through a trained feature extraction model to obtain sensitive privacy features under at least one sensitive information category; then, non-numerical data in each extracted sensitive privacy feature is converted into numerical data through a preset encoding method to obtain each sensitive privacy feature value corresponding to the privacy sample data.

[0057] As for the feature extraction model and encoding method used here, they can be flexibly set based on actual needs and are not specifically limited in this embodiment.

[0058] Step 202: normalize each sensitive privacy feature value corresponding to the privacy sample data to obtain a normalized processing result corresponding to the privacy sample data.

[0059] In this embodiment, the normalization process refers to scaling the data to a specified range, where the specified range may be, for example, a range greater than or equal to 0 and less than or equal to 1.

[0060] Step 203, based on the correspondence between the number of sensitive privacy feature values ​​and the number of nodes on the model input layer, and in accordance with a specified dimensionality reduction method, a dimensionality reduction process is performed on the normalized processing result corresponding to the privacy sample data to obtain a sensitive privacy feature vector corresponding to the privacy sample data.

[0061] In this embodiment, the normalization processing result corresponding to the privacy sample data refers to the sensitive privacy feature value after normalization processing corresponding to the privacy sample data. The specified dimensionality reduction method can be, for example, a nonlinear data dimensionality reduction method such as Kernel Principal Component Analysis (KPCA), which is not specifically limited in this embodiment.

[0062] As described above, the number of sensitive privacy feature values ​​corresponding to the number of nodes on the model input layer means that the number of sensitive privacy feature values ​​is less than or equal to the number of nodes on the model input layer. Based on this, this step 203 can be implemented as follows: based on the premise that the number of sensitive privacy feature values ​​corresponds to the number of nodes on the model input layer, first traverse the current normalized processing result, take the traversed sensitive privacy feature value as the current sensitive privacy feature value, calculate the feature correlation between the current sensitive privacy feature value and other sensitive privacy feature values ​​in the current normalized processing result except the sensitive privacy feature value according to the feature correlation algorithm indicated by the specified dimensionality reduction method, then remove at least one other sensitive privacy feature value whose feature correlation with the current sensitive privacy feature value is greater than the set correlation threshold from the current normalized processing result, and continue to traverse the normalized processing result. After the traversal is completed, a sensitive privacy feature vector corresponding to the privacy sample data is constructed based on the sensitive privacy feature value in the current normalized result, where the number of sensitive privacy feature values ​​in the constructed sensitive privacy feature vector corresponding to the privacy sample data corresponds to the number of nodes on the model input layer.

[0063] For example, taking the port number and data transmission protocol type used by the device, since the data transmission protocol type used determines which port number the device uses, in actual application, you only need to know one of them to know the other, so only one of them needs to be retained; that is, the sensitive privacy feature value corresponding to the port number used by the device and the sensitive privacy feature value corresponding to the data transmission protocol type can be considered as sensitive privacy feature values ​​with a higher feature correlation, so in actual application, you only need to retain at least one of the sensitive privacy feature values ​​with a higher feature correlation.

[0064] In this embodiment, as an embodiment, based on the correspondence between the number of sensitive privacy feature values ​​and the number of nodes on the model input layer, and before performing dimensionality reduction processing on the normalized processing results corresponding to each privacy sample data according to a specified dimensionality reduction method, redundant removal processing may be performed on the normalized processing results corresponding to each privacy sample data to obtain redundant removal processing results corresponding to each privacy sample data. In this way, redundant data in each privacy sample data can be removed, the quality of the model training data can be improved, and the training performance of the model can be improved during subsequent model training.

[0065] Specifically, this embodiment can first obtain the current privacy sample data set, and the normalization processing result corresponding to any privacy sample data in the current privacy sample data set has not been removed as redundant data; traverse the current privacy sample data set, and use the currently traversed privacy sample data as the current privacy sample data, and then calculate the similarity between the normalization processing result corresponding to the current privacy sample data and the normalization processing results corresponding to other privacy sample data in the current privacy sample data set except the current privacy sample data through a preset similarity algorithm, and remove the normalization processing results corresponding to the privacy sample data with a similarity greater than a set similarity threshold as redundant data, and continue to traverse the current privacy sample data set. After the traversal is completed, the normalization processing results corresponding to each privacy sample data in the current privacy sample data set are used as the redundant removal processing results corresponding to the above-mentioned privacy sample data.

[0066] The above step 103 is further described below:

[0067] In this embodiment, as an embodiment, there are many specific implementation methods for selecting the DNN models that match the M CPU cores respectively from the DNN models corresponding to the N CPU cores in the above step 103, for example, see Figure 3 As shown, the specific implementation may include the following steps:

[0068] Step 301, for each CPU core among the M CPU cores, obtain a current DNN model set, wherein any DNN model in the current DNN model set is not assigned to any CPU core among the M CPU cores.

[0069] Step 302, traverse the current DNN model set, take the traversed DNN model as the current DNN model, determine whether the CPU core that trained the current DNN model is the same as the architecture of the CPU core in the M CPU cores, if yes, take the current DNN model as a candidate DNN model, and continue to traverse the current DNN model set, if not, continue to traverse the current DNN model set.

[0070] Step 303: Select a DNN model with the highest model accuracy from each candidate DNN model as the DNN model that matches the CPU core among the M CPU cores.

[0071] In this embodiment, the detection result in the above step 103 indicates the predicted probability value that the target network flow belongs to at least one configured sensitive information category. Based on this, as an embodiment, the target prediction result is determined based on M detection results in the above step 103. In the specific implementation, for example, it can be: first for each detection result in the M detection results, if the detection result includes at least one predicted probability value greater than or equal to the set probability value, it means that the detection result indicates that the above target network flow has privacy leakage, and the detection result can be used as the first reference result at this time; otherwise, it means that the detection result indicates that the above target network flow does not have privacy leakage, and the detection result can be used as the second reference result at this time; thereafter, if the number of detection results used as the first reference result in the M detection results is greater than or equal to the number of detection results used as the second reference result, it is determined that the target prediction result indicates that the target network flow has privacy leakage, otherwise, it is determined that the target prediction result indicates that the target network flow has no privacy leakage.

[0072] This embodiment uses multiple trained DNN models to perform privacy leakage detection on the target network flow, so as to comprehensively judge whether there is privacy leakage in the target network flow based on the multiple detection results obtained, which can effectively improve the accuracy of privacy leakage detection of the network flow.

[0073] At this point, the description of the method provided in the embodiment of the present application is completed. The following is a description of the device provided in the embodiment of the present application:

[0074] As an embodiment, this embodiment also provides a privacy leakage detection device based on a parallel deep neural network. Figure 4 , Figure 4 A schematic diagram of the device structure provided in an embodiment of the present application.

[0075] like Figure 4 As shown, the device 400 is applied to an electronic device, and the device 400 includes: an acquisition module 401, an extraction module 402 and a training module 403;

[0076] The obtaining module 401 is used to obtain a privacy data set; the privacy data set includes privacy sample data corresponding to multiple network flows respectively; any network flow is initiated when a business function with privacy data is triggered;

[0077] An extraction module 402 is used to extract at least one sensitive privacy feature under a sensitive information category from each privacy sample data in the privacy data set to obtain a sensitive privacy feature vector corresponding to the privacy sample data;

[0078] The training module 403 is used to control N CPU cores on the device to train the corresponding DNN model in parallel based on the sensitive privacy feature vector corresponding to each privacy sample data; in the DNN model corresponding to any CPU core, the number of nodes on the input layer corresponds to the number of eigenvalues ​​in the sensitive privacy feature vector, and the number of nodes on the output layer corresponds to the number of sensitive information categories; different CPU cores use different optimizers for training the DNN model; N is greater than 1;

[0079] Among them, the DNN models corresponding to the N CPU cores are used. When any detection device needs to detect any target network flow, the detection device selects the DNN models that match the M CPU cores respectively from the DNN models corresponding to the N CPU cores based on the M idle CPU cores in the detection device, and the M CPU cores use the matching DNN models to perform IoT privacy leakage detection on the target network flow, and determine the target prediction result based on the M detection results; the target prediction result indicates whether there is privacy leakage in the target network flow, M is greater than 1 and less than or equal to N; when M is less than N, the DNN model matched by any CPU core is trained by a CPU core with the same architecture as the CPU core and has the highest model accuracy among all the remaining DNN models.

[0080] As an embodiment, the extraction module 402 is specifically used for:

[0081] For each privacy sample data in the privacy data set, extract at least one sensitive privacy feature under the sensitive information category from the privacy sample data, and convert the non-numerical data in each extracted sensitive privacy feature into numerical data to obtain each sensitive privacy feature value corresponding to the privacy sample data;

[0082] Normalizing each sensitive privacy feature value corresponding to the privacy sample data to obtain a normalized processing result corresponding to the privacy sample data;

[0083] Based on the correspondence between the number of sensitive privacy feature values ​​and the number of nodes on the model input layer, the normalized processing result corresponding to the privacy sample data is reduced in dimension according to the specified dimensionality reduction method to obtain the sensitive privacy feature vector corresponding to the privacy sample data.

[0084] As an embodiment, the number of feature values ​​in the sensitive privacy feature vectors corresponding to the privacy sample data is the same;

[0085] The number of nodes on the input layer of the DNN model corresponding to N CPU cores is the same;

[0086] The number of nodes on the output layer in the DNN model corresponding to N CPU cores is the same.

[0087] As an embodiment, selecting the DNN models that match the M CPU cores respectively from the DNN models corresponding to the N CPU cores includes:

[0088] For each CPU core in the M CPU cores, a current DNN model set is obtained, and any DNN model in the current DNN model set is not assigned to any CPU core in the M CPU cores; the current DNN model set is traversed, and the traversed DNN model is used as the current DNN model, and it is determined whether the CPU core for training the current DNN model is the same as the architecture of the CPU core in the M CPU cores. If yes, the current DNN model is used as a candidate DNN model, and the current DNN model set is continued to be traversed. If not, the current DNN model set is continued to be traversed;

[0089] The DNN model with the highest model accuracy is selected from each candidate DNN model as the DNN model matching the CPU core among the M CPU cores.

[0090] As an embodiment, the detection result indicates a predicted probability value that the target network flow belongs to at least one configured sensitive information category;

[0091] Determining the target prediction result based on M detection results includes:

[0092] For each of the M test results, if the test result includes at least one predicted probability value greater than or equal to the set probability value, the test result is used as the first reference result; otherwise, the test result is used as the second reference result;

[0093] If the number of detection results used as the first reference results among the M detection results is greater than or equal to the number of detection results used as the second reference results, then it is determined that the target prediction result indicates that privacy leakage occurs in the target network flow; otherwise, it is determined that the target prediction result indicates that no privacy leakage occurs in the target network flow.

[0094] As an embodiment, at least two CPU cores among the N CPU cores have different architectures.

[0095] So far, completed Figure 4 Structural description of the device shown.

[0096] The implementation process of the functions and effects of each module in the above-mentioned device is specifically described in the implementation process of the corresponding steps in the above-mentioned method, which will not be repeated here.

[0097] For the device embodiment, since it basically corresponds to the method embodiment, the relevant parts can refer to the partial description of the method embodiment. The device embodiment described above is only schematic, wherein the modules described as separate components may or may not be physically separated, and the components displayed as modules may or may not be physical modules, that is, they may be located in one place, or they may be distributed on multiple network modules. Some or all of the modules may be selected according to actual needs to achieve the purpose of the present application scheme. A person of ordinary skill in the art can understand and implement it without paying creative labor.

[0098] See also Figure 5 , is a schematic diagram of the hardware structure of an electronic device provided by an exemplary embodiment of the present application. The electronic device may include a processor 501, a communication interface 502, a computer-readable storage medium 503 and a communication bus 504. The processor 501, the communication interface 502 and the computer-readable storage medium 503 communicate with each other through the communication bus 504. Among them, computer program instructions are stored on the computer-readable storage medium 503; the processor 501 can execute the steps of the method described in the above embodiment by executing the computer program instructions stored on the computer-readable storage medium 503. The electronic device may also include other hardware according to the actual function of the electronic device, which will not be described in detail.

[0099] Correspondingly, an embodiment of the present application further provides a computer-readable storage medium, on which a number of computer program instructions are stored. When the computer program instructions are executed by a processor, the method disclosed in the above example of the present application can be implemented.

[0100] Exemplarily, the computer-readable storage medium may be any electronic, magnetic, optical or other physical storage device that may contain or store information, such as executable instructions, data, etc. For example, the computer-readable storage medium may be: RAM (Radom Access Memory), volatile memory, non-volatile memory, flash memory, storage drive (such as hard disk drive), solid state drive, any type of storage disk (such as optical disk, DVD, etc.), or similar storage medium, or a combination thereof. The processor and memory may be supplemented by or incorporated into a dedicated logic circuit.

[0101] The above are only preferred embodiments of the present application and are not intended to limit the present application. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present application should be included in the scope of protection of the present application.

Claims

1. A privacy leakage detection method based on parallel deep neural network, characterized in that: The method is applied to an electronic device, and the method comprises: Obtaining a privacy data set; the privacy data set includes privacy sample data corresponding to a plurality of network flows respectively; any network flow is initiated when a business function with privacy data is triggered; Extracting at least one sensitive privacy feature under a sensitive information category from each privacy sample data of the privacy data set to obtain a sensitive privacy feature vector corresponding to the privacy sample data; Based on the sensitive privacy feature vector corresponding to each privacy sample data, control the N CPU cores of the device to train the corresponding deep neural network DNN model in parallel; in the DNN model corresponding to any CPU core, the number of nodes on the input layer corresponds to the number of eigenvalues ​​in the sensitive privacy feature vector, and the number of nodes on the output layer corresponds to the number of sensitive information categories; different CPU cores use different optimizers for training the DNN model; N is greater than 1; Among them, the DNN models corresponding to the N CPU cores are used for, when any detection device needs to detect any target network flow, the detection device selects the DNN models that match the M CPU cores respectively from the DNN models corresponding to the N CPU cores based on the M idle CPU cores in the detection device, and the M CPU cores respectively use the matching DNN models to perform IoT privacy leakage detection on the target network flow, and determine the target prediction result based on the M detection results; the target prediction result indicates whether there is privacy leakage in the target network flow, and M is greater than 1 and less than or equal to N; when M is less than N, the DNN model matched by any CPU core is trained by a CPU core with the same architecture as the CPU core and has the highest model accuracy among all the remaining DNN models.

2. The method according to claim 1, characterized in that The step of extracting at least one sensitive privacy feature under a sensitive information category from each privacy sample data of the privacy data set to obtain a sensitive privacy feature vector corresponding to the privacy sample data includes: For each privacy sample data of the privacy data set, extract at least one sensitive privacy feature under the sensitive information category from the privacy sample data, and convert non-numerical data in each extracted sensitive privacy feature into numerical data to obtain each sensitive privacy feature value corresponding to the privacy sample data; Normalizing each sensitive privacy feature value corresponding to the privacy sample data to obtain a normalized processing result corresponding to the privacy sample data; Based on the correspondence between the number of sensitive privacy feature values ​​and the number of nodes on the model input layer, the normalized processing result corresponding to the privacy sample data is reduced in dimension according to the specified dimensionality reduction method to obtain the sensitive privacy feature vector corresponding to the privacy sample data.

3. The method according to claim 1, characterized in that The number of eigenvalues ​​in the sensitive privacy feature vector corresponding to each privacy sample data is the same; The number of nodes on the input layer of the DNN model corresponding to N CPU cores is the same; The number of nodes on the output layer in the DNN model corresponding to N CPU cores is the same.

4. The method according to claim 1, characterized in that The selecting the DNN models that match the M CPU cores respectively from the DNN models corresponding to the N CPU cores includes: For each CPU core in the M CPU cores, a current DNN model set is obtained, wherein any DNN model in the current DNN model set is not assigned to any CPU core in the M CPU cores; the current DNN model set is traversed, and the traversed DNN model is used as the current DNN model, and it is determined whether the CPU core trained with the current DNN model has the same architecture as the CPU core in the M CPU cores, if yes, the current DNN model is used as a candidate DNN model, and the current DNN model set is continued to be traversed, and if not, the current DNN model set is continued to be traversed; A DNN model with the highest model accuracy is selected from each candidate DNN model as the DNN model that matches the CPU core among the M CPU cores.

5. The method according to claim 1, characterized in that The detection result indicates a predicted probability value that the target network flow belongs to at least one configured sensitive information category; Determining the target prediction result based on the M detection results includes: For each of the M test results, if the test result includes at least one predicted probability value greater than or equal to a set probability value, the test result is used as a first reference result; otherwise, the test result is used as a second reference result; If the number of detection results serving as the first reference results among the M detection results is greater than or equal to the number of detection results serving as the second reference results, it is determined that the target prediction result indicates that a privacy leak has occurred in the target network flow; otherwise, it is determined that the target prediction result indicates that no privacy leak has occurred in the target network flow.

6. The method according to claim 1, characterized in that At least two CPU cores among the N CPU cores have different architectures.

7. A privacy leakage detection device based on parallel deep neural network, characterized in that: The device is applied to electronic equipment, and the device comprises: An acquisition module is used to obtain a private data set; the private data set includes private sample data corresponding to a plurality of network flows respectively; any network flow is initiated when a business function with private data is triggered; An extraction module, configured to extract at least one sensitive privacy feature under a sensitive information category from each privacy sample data of the privacy data set, so as to obtain a sensitive privacy feature vector corresponding to the privacy sample data; A training module, used for controlling N CPU cores on the device to train corresponding deep neural network DNN models in parallel based on the sensitive privacy feature vectors corresponding to each privacy sample data; in the DNN model corresponding to any CPU core, the number of nodes on the input layer corresponds to the number of eigenvalues ​​in the sensitive privacy feature vector, and the number of nodes on the output layer corresponds to the number of sensitive information categories; different CPU cores use different optimizers for training the DNN model; N is greater than 1; Among them, the DNN models corresponding to the N CPU cores are used for, when any detection device needs to detect any target network flow, the detection device selects the DNN models that match the M CPU cores respectively from the DNN models corresponding to the N CPU cores based on the M idle CPU cores in the detection device, and the M CPU cores respectively use the matching DNN models to perform IoT privacy leakage detection on the target network flow, and determine the target prediction result based on the M detection results; the target prediction result indicates whether there is privacy leakage in the target network flow, and M is greater than 1 and less than or equal to N; when M is less than N, the DNN model matched by any CPU core is trained by a CPU core with the same architecture as the CPU core and has the highest model accuracy among all the remaining DNN models.

8. The device according to claim 7, characterized in that The extraction module is specifically used to: for each privacy sample data of the privacy data set, extract at least one sensitive privacy feature under the sensitive information category from the privacy sample data, convert the non-numerical data in each extracted sensitive privacy feature into numerical data to obtain each sensitive privacy feature value corresponding to the privacy sample data; perform normalization processing on each sensitive privacy feature value corresponding to the privacy sample data to obtain a normalized processing result corresponding to the privacy sample data; based on the correspondence between the number of sensitive privacy feature values ​​and the number of nodes on the model input layer, perform dimensionality reduction processing on the normalized processing result corresponding to the privacy sample data according to a specified dimensionality reduction method to obtain a sensitive privacy feature vector corresponding to the privacy sample data; and / or, The number of eigenvalues ​​in the sensitive privacy feature vector corresponding to each privacy sample data is the same; the number of nodes on the input layer of the DNN model corresponding to the N CPU cores is the same; the number of nodes on the output layer of the DNN model corresponding to the N CPU cores is the same; and / or, The selecting of the DNN models that match the M CPU cores respectively from the DNN models corresponding to the N CPU cores includes: obtaining a current DNN model set for each CPU core in the M CPU cores, wherein any DNN model in the current DNN model set is not assigned to any CPU core in the M CPU cores; traversing the current DNN model set, taking the traversed DNN model as the current DNN model, and determining whether the CPU core that trains the current DNN model has the same architecture as the CPU core in the M CPU cores; if so, taking the current DNN model as a candidate DNN model and continuing to traverse the current DNN model set; if not, continuing to traverse the current DNN model set; selecting a DNN model with the highest model accuracy from each candidate DNN model as the DNN model that matches the CPU core in the M CPU cores; and / or, The detection result indicates a predicted probability value that the target network flow belongs to at least one configured sensitive information category; determining the target prediction result based on M detection results includes: for each detection result among the M detection results, if the detection result includes at least one predicted probability value greater than or equal to a set probability value, then the detection result is used as a first reference result, otherwise the detection result is used as a second reference result; if the number of detection results used as first reference results among the M detection results is greater than or equal to the number of detection results used as second reference results, then determining that the target prediction result indicates that privacy leakage occurs in the target network flow, otherwise, determining that the target prediction result indicates that no privacy leakage occurs in the target network flow; and / or, At least two CPU cores among the N CPU cores have different architectures.

9. An electronic device, characterized in that: The electronic device includes: Processor; and A computer-readable storage medium, wherein computer program instructions are stored in the computer-readable storage medium, and when the computer program instructions are executed by the processor, the processor executes the steps in any one of the methods of claims 1 to 6.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer program instructions, which, when executed by a processor, enable the processor to perform the steps of any one of the methods of claims 1 to 6.

Images